TFEA runtime and output differs depending on order of conditions

[santiago-es]

Hello all, first off thank you for your hard work providing this tool to the community! It's a very useful, interesting addition.

I've attempted to apply the pipeline to our ATAC-seq data on an experiment with two conditions, A and B with two bioreps per condition. I opted for the config file input type. I first ran Condition 1 = A and Condition 2 = B, plotall=TRUE completing the pipeline in ~7-8 hours. I then attempted Condition 1 = B, Condition 2= A plotall=FALSE and the run completed in 2 days and 6 hours on the same machine. I expected the output to be identical with opposite signs for the calculated E-score on the respective transcription factors, however the outputs differed significantly as did the p-values.

I am copying both config and results files here.
tfea_config_Run1.txt
tfea_config_Run1.txt
results_Run1.txt
results_Run2.txt

To aid in comparing the different results, I've copied the top of the results.html tables here.
Run 1:

Run 2: (note: the expectation is the identical table from above with the positive E-score table in Run 2 matching the negative E-score table in Run 1, and vice versa).

I'd greatly appreciate any help understanding this issue, I'm still a novice when it comes to bioinformatics.

[jtstanley]

Hi @santiago-es thank you for the feedback. We're happy to hear you're making use of TFEA. Pardon the delay on responding, the ownership of the code base has been in flux.

In principle, we don't necessarily expect the E-score to be the exact value up to the sign when you switch condition labels. Though this will exactly flip the ordering of the regions, the way the weighting scheme for the background distribution is calculated is not necessarily symmetric. However, the big change in adjusted p-value seems suspect. We'll look more at your results files and get back to you with more questions. Thanks for your patience!